The SME AI adoption gap: usage is high, maturity is low.
The OECD's 2026 D4SME Survey and its December 2025 paper for the G7 approach SME artificial intelligence from two angles: use is broad, structured learning is rare. The practices associated with higher maturity are ordinary decisions about time and training.
The OECD published two documents on small firms and artificial intelligence within four months of each other. The first, a discussion paper written for the G7 ministerial meeting in Montreal on 7–9 December 2025, fed the proposed G7 SME AI Adoption Blueprint, whose six action areas include skilling and business strategy. The second, Empowering SMEs in the age of AI, published in April 2026, reports the annual Digital for SMEs survey: 2,018 responses gathered between the fourth quarter of 2025 and the first quarter of 2026.
One number from the survey has travelled widely: 61% of responding firms use at least one AI-enabled application, while 76% of those users are classified as “AI novices”, applying standard tools to isolated tasks. A figure in the December paper bears on the distance between the two. In every G7 country covered by a 2024 OECD survey of SMEs using generative AI, under 30% report that their employees take part in any AI-related training: 29.4% in Canada, 24.0% in the United Kingdom, 23.2% in Germany, 11.3% in Japan.
The tools have diffused. The capacity to use them has not, and that is a management question before a technical one.
The survey that produced the adoption figure also recorded how firms acquire skills
The sample deserves stating plainly. Responses came from twelve countries, 1,376 of the 2,018 from Japan alone, so results are reported as country averages. Firms were reached through the digital platforms on which they trade, and the OECD states that the sample is neither randomised nor representative of national SME populations. Switzerland is not among the twelve; Swiss firms have to be read in from other evidence.
The survey also asked how businesses address their digital skills needs. The most common answers were internet search (32%), external consultants (29%), friends and family (24%) and generative AI itself (21%). The OECD’s summary: skills are sourced through informal and fragmented pathways.
Against that, the strategy figures look confident: 70% of AI users report a defined approach to AI integration, 60% of all respondents a digitalisation strategy. Yet 22% of firms sit at the “basic” level of digital maturity and three quarters of AI users remain novices. A strategy is easily declared; what most firms actually do to build the capability behind it is search the internet.
Structured training is what distinguishes the more mature firms
The clearest evidence sits in the Japanese sub-sample, large enough for regression analysis on 1,376 responses. Three practices are associated with higher digital maturity: stronger cybersecurity measures, a defined digitalisation approach rather than none, and — the finding that matters here — addressing skills needs through structured training programmes, for instance with specialist institutions, or through internal capacity building such as in-house training and mutual learning. Informal approaches, including ad hoc online resources and advice from friends and family, show no clear association. The OECD calls these conditional associations, not causal effects, in a sample that is not representative.
The December paper points the same way, citing evidence that firm-provided training significantly increases workers’ use of generative AI. The difference is made by what the firm organises, not by what employees improvise.
The productivity advantage of AI users shrinks once capabilities are taken into account
There is a productivity premium associated with AI use: comparing firms of similar size, age and sector across G7 countries, the OECD finds premia in most cases above 4% and sometimes above 15%. Two qualifications come with it. Part of the premium reflects selection, since firms that were already more digital and more competitive adopt AI first. And once connectivity, digital capabilities such as cloud computing and workers’ ICT skills — proxied by ICT specialists or relevant training — are accounted for, the measured advantage shrinks substantially.
Returns depend on complementary investments and on integration into operations, which is why productivity can dip before it rises. The technology is not the asset; the capability built around it is. Among SMEs using generative AI, only 29% use it in their core activities.
Expertise and time are what firms say they lack, and Swiss structure tightens both
Europe’s official statistics record what firms that have not adopted actually say. In 2025, 20.0% of EU enterprises with ten or more employees used AI technologies, up from 13.5% in 2024: 17% of small firms against 55.0% of large ones. Among those that considered AI and decided against it, the most cited reason was lack of relevant expertise (70.9%), ahead of unclear legal consequences (52.5%) and data protection concerns (48.8%). Only 20.7% said the technologies were not useful for their business.
Asked about barriers to digital adoption, D4SME respondents named maintenance costs (39%), lack of time for training (38%), hardware costs (37%) and the cost of training (23%); regulatory complexity came far behind, at 13%. Across OECD countries, work-related time constraints are the most cited barrier to job-related non-formal learning, and smaller firms have less flexibility to release staff from revenue-generating work, face higher training costs per worker, and hesitate when trained staff can be recruited away.
Switzerland’s structure concentrates that constraint. In the Federal Statistical Office’s business structure statistics for 2023, the country counted 626,033 market-economy enterprises, of which 624,219 — more than 99% — employed fewer than 250 people, together providing two thirds of its 4.82 million jobs. Micro-enterprises of one to nine people account for 561,952 of those firms and 1.19 million persons employed, an average of around two per business. In a company of that size a two-day course is a measurable share of monthly capacity, with no training function to organise it. Sourcing skills by internet search is not negligence; it is what remains when nobody has been given the hours.
Swiss survey evidence shows the familiar shape. The SME labour-market study published by AXA Switzerland on 8 October 2025 — Sotomo, 300 SMEs in German- and French-speaking Switzerland, fieldwork in March 2025 — found 34% deliberately integrating AI into work processes, up from 22% a year earlier, and 37% testing it; 57% of AI-using firms reported time savings, against 46% the year before, while only about a third had clear data-protection rules for AI use, falling to 23% in firms with five to nine employees. A broader survey published by EY Switzerland on 27 May 2026, covering 604 people working in Swiss companies and weighted towards large employers, found 89% using AI in their working day but 9% reporting a change to the business model.
What a management team can decide before buying the next licence
Nothing here suggests that firms reporting deeper value bought better technology. They organised learning — a short list of decisions open to any management team.
- Book the hours before the tool: name the process, the people, and the hours per month for which they are released from other work. Structured programmes and internal capacity building are what correlate with maturity; ad hoc searching is not.
- Measure one recurring process before and after. The survey reports perceptions, not performance: 54% of AI users see at least moderate benefits, 21% significant or transformational ones. Only your own count of hours or error rates will say which you have.
- Make the learning internal: someone owns the practice, writes down what works and teaches the next person. Documented know-how and mutual learning are among the practices associated with higher maturity.
- Check what public support exists: 16% of surveyed SMEs had used a public digitalisation programme, and 65% of the rest reported not knowing such programmes existed.
- Treat AI literacy as an obligation wherever operations fall under European Union law. Article 4 of the AI Act, applicable since 2 February 2025 and amended in July 2026 by the Digital Omnibus on AI, requires providers and deployers to take measures supporting the development of AI literacy among staff and others operating AI on their behalf, with no specific level mandated. National market surveillance authorities have supervised the rules since 2 August 2026, and the Commission’s repository lists more than 40 documented practices.
Two series will show whether this moves: the annual D4SME survey, and Eurostat’s enterprise ICT statistics, whose 2025 results appeared on 11 December 2025. Until then, the question a board can answer for itself is not how many AI tools the company has bought, but how many hours of structured training it scheduled last quarter, and which process it measured before and after.